ABSTRAKSI: Head tracking merupakan suatu proses penelusuran dan pendeteksian pose dan pergerakan kepala oleh sistem berbasis komputer dengan menggunakan inputan video yang selanjutnya digunakan untuk mendukung kebutuhan aplikasi pada bidang tertentu. Misalnya diterapkan dalam bidang Computer Vision, Intelligence vehicle, Multimedia, dsb. Dengan semakin maraknya aplikasi yang membutuhkan implementasi head tracking sebagai faktor pendukung sistem tersebut, timbul suatu permasalahan, yakni bagaimana mengimplementasikan head tracking dengan baik.
Dalam tugas akhir ini, untuk membangun sistem yang bisa mengimplementasikan head tracking, digunakan metode Active Appearance Model (AAM) dengan algoritma fitting Simultaneous Inverse Compositonal (SIC). AAM akan menghasilkan suatu model yang bersifat deformable dimana bisa berubah dan menyesuaikan serta mengestimasi pose dari citra input (frameframe video) melalui proses fitting. Model dibangun melalui proses training dari sekumpulan citra latih sehingga dihasilkan dua representasi model, yakni shape dan appearance yang memiliki variasi atau mode masing-masing. Jumlah mode dari model menentukan performansi model dalam proses fitting. Selain itu, inisialisasi penempatan model juga mempengaruhi perfomansi proses fitting dalam mencapai konvergensi.
Dari hasil pengujian implementasi head tracking pada dataset utama dapat diketahui bahwa metode AAM dengan penentuan parameter-parameter yang terbaik mampu mengimplementasikan head tracking dengan akurasi sistem sebesar 95.33%.Kata Kunci : head tracking, Active Appearance Model, Simultaneous Inverse Compositional, Shape Model, Appearance ModelABSTRACT: Head tracking is a head movement tracking and detecting process by computerized based system using video input which is used to support other needs in each kind of applications. For example in supporting an application of computer vission, intelligence vehicle, multimedia and others. Because of many applications which needs head tracking implementation in supporting their main main functions, it was emerging some problem, such as how to implement a good head tracking process.
In this final project, the method used to implement head tracking is Active Appearance Model (AAM) combining Simultaneous Inverse Compositional (SIC) as a fitting algorithm. AAM will build deformable model which can change, fit and estimate input image (video frames) pose using fitting process. That model is built by using training process from a set of training images which is resulting two representative model, called shape and appearance. Both of them has their own variations or modes of movement which can influence model performance in fitting process. On the other hand, the model initial estimation process influence the fitting process too in reaching convergence.
From the result of the experiment in implementing head tracking on main dataset, can be known that AAM method using the best parameter approximation could implemented head tracking with 95.33% system accuration.Keyword: head tracking, Active Appearance Model, Simultaneous Inverse Compositional, Shape Model, Appearance Model